[srslte-users] srsenb&ue running on ubuntu vs centos

Fri Mar 26 17:07:04 UTC 2021

Interesting.
I put the file .srslte_fftwisdom from the host without delay to the new one, and the srsue startup delay has gone somewhere. Magic)

Thanks!

-----Original Message-----
From: srslte-users <srslte-users-bounces at lists.softwareradiosystems.com> On Behalf Of J Giovatto
Sent: Friday, March 26, 2021 4:04 PM
To: srslte-users at lists.softwareradiosystems.com
Subject: Re: [srslte-users] srsenb&ue running on ubuntu vs centos

Here is an examle on fedora32 that started w/o delay.

sudo strace -f ./srsue ./ue.conf.example 2<&1 | tee foo.log | grep fftw

openat(AT_FDCWD, "/usr/local/lib64/libfftw3f.so.3", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, "/home/jgiovatto/Devel/LTE/srs/srsLTE-master_2020_10/build/lib/src/phy/rf/libfftw3f.so.3",
O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, "tls/haswell/x86_64/libfftw3f.so.3",
O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, "tls/haswell/libfftw3f.so.3", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, "tls/x86_64/libfftw3f.so.3", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, "tls/libfftw3f.so.3", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, "haswell/x86_64/libfftw3f.so.3", O_RDONLY|O_CLOEXEC) =
-1 ENOENT (No such file or directory)
openat(AT_FDCWD, "haswell/libfftw3f.so.3", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, "x86_64/libfftw3f.so.3", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, "libfftw3f.so.3", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, "/lib64/libfftw3f.so.3", O_RDONLY|O_CLOEXEC) = 3 openat(AT_FDCWD, "/root/.srslte_fftwisdom", O_RDONLY) = 3 read(3, "(fftw-3.3.8 fftwf_wisdom #x9e7d4"..., 4096) = 4096 read(3, "ac4b #x7456e959)\n  (fftwf_rdft_r"..., 4096) = 4096 read(3, "50 #xf829999d)\n  (fftwf_codelet_"..., 4096) = 4096

On 3/26/21 6:55 AM, alexander.serkin at gmail.com wrote:
> Strace output:
> ...
> fcntl(14, F_SETFL, O_RDONLY|O_NONBLOCK) = 0 write(1, "Waiting PHY to 
> initialize ... ", 30Waiting PHY to initialize ... ) = 30 
> futex(0x7fe2806f2060, FUTEX_WAIT_PRIVATE, 2, NULL) ###! The delay is 
> here !###  = 0 futex(0x7fe2806f2060, FUTEX_WAKE_PRIVATE, 1) = 0 
> write(1, "done!\n", 6done!
>
> And the problem persists after reinstalling the VM from centos back to ubuntu 18.
>
> -----Original Message-----
> From: alexander.serkin at gmail.com <alexander.serkin at gmail.com>
> Sent: Friday, March 26, 2021 1:44 PM
> To: 'Cedric Roux' <cedric.roux at eurecom.fr>; 'Andre Puschmann' 
> <andre.puschmann at softwareradiosystems.com>; 
> srslte-users at lists.softwareradiosystems.com
> Subject: RE: [srslte-users] srsenb&ue running on ubuntu vs centos
>
> Hi Cedric
> Not sure I did all correct, but I changed rrc.cc that way:
> ..
> #include <random>
> ..
> using namespace std;
> ..
>   // set seed (used in CHAP auth and attach)
>   // srand(tv.tv_usec);
>   mt19937 mt_rand(tv.tv_usec);
> ...
>
> And that did not help - the delay seems to occur somewhere later on after the call to srand or mt_rand. And even after the rrc init procedure:
>
> 10:41:33.131890 [NAS ] [I] Read security ctxt from file .ctxt. ksi: 5, 
> k_asme: 
> a99f174aa53110ed5020762235c7ccf7012099cc8406e2a68a9472b201cdb9a8, 
> tx_count: 4, rx_count: 3, int_alg: 1, enc_alg: 0
> 10:41:33.131935 [RRC ] [I] using mt_rand of 131931
> 10:41:33.131943 [RRC ] [I] mt_rand ended. starting 
> ue_required_sibs.assign
> 10:41:33.131944 [RRC ] [I] ue_required_sibs.assign ended Ooops...
> 10:41:58.497617 [MAC ] [I] [   11] mac::get_metrics started
>
> Cheers,
> Alex
>
> -----Original Message-----
> From: Cedric Roux <cedric.roux at eurecom.fr>
> Sent: Friday, March 26, 2021 12:48 PM
> To: Andre Puschmann <andre.puschmann at softwareradiosystems.com>; 
> alexander.serkin at gmail.com; 
> srslte-users at lists.softwareradiosystems.com
> Subject: Re: [srslte-users] srsenb&ue running on ubuntu vs centos
>
>
>
> On 2021-03-25 16:19, Andre Puschmann wrote:
>> Hey Alex,
>>
>> On 24/3/21 10:10, alexander.serkin at gmail.com wrote:
>>> Despite the fact that centos VM has twice more cpus and 8 times more 
>>> RAM comparing to ubuntu, the srsue starts there >20 seconds while on 
>>> ubuntu machine it takes 2-3 seconds.
>> we've actually observed the same on some configuration. Yet, I've not 
>> come to find the cause. I was suspecting the srand() which we should
> maybe there is an access to /dev/random?
> Reading data from /dev/random can be slow, based on the activity of the computer.
> You can check with strace maybe.
>
>> probably not use anyhow and instead port to C++ std::mt19937, like we 
>> already do elsewhere in the code
>>
>> If you can reliably reproduce that, try to change to std::mt19937 and 
>> see if that improves the start up time.
>>
>> Thanks
>> Andre
>>
>>> The delay occurs between two UE log messages. CentOS:
>>>
>>> 08:55:21.243251 [NAS ] [I] Read security ctxt from file .ctxt. ksi: 
>>> 6,
>>> k_asme:
>>> b03a9dc5106b1c09e96b14a11ed34afe128d632bea9509b902fadeae01669ce3,
>>> tx_count: 4, rx_count: 3, int_alg: 1, enc_alg: 0
>>>
>>> 08:55:21.243307 [RRC ] [I] using srand seed of 243303
>>>
>>> 08:55:40.753622 [MAC ] [I] [    0] DL retx: 0.00 %, perpkt: 0.00, UL
>>> retx: 0.00 % perpkt: 0.00
>>>
>>> 08:55:40.753726 [RLC ] [I] LCID=0, RX throughput: 0.000000 Mbps. TX
>>> throughput: 0.000000 Mbps.
>>>
>>>   
>>>
>>> Ubuntu:
>>>
>>> 08:58:46.389003 [NAS ] [I] Read security ctxt from file .ctxt. ksi: 
>>> 6,
>>> k_asme:
>>> 0417d4c4d86b4d272a061cc3ae49f177722fe64870e7a7aec63d47033162be6d,
>>> tx_count: 14, rx_count: b, int_alg: 1, enc_alg: 0
>>>
>>> 08:58:46.389025 [RRC ] [I] using srand seed of 389022
>>>
>>> 08:58:47.075794 [MAC ] [I] [    0] DL retx: 0.00 %, perpkt: 0.00, UL
>>> retx: 0.00 % perpkt: 0.00
>>>
>>> 08:58:47.075809 [RLC ] [I] LCID=0, RX throughput: 0.000000 Mbps. TX
>>> throughput: 0.000000 Mbps.
>>>
>>>   
>>>
>>> I wonder if there probably exist some simple explanation of this ?
>>>
>>>   
>>>
>>> Cheers,
>>>
>>> Alex
>>>
>>>
>>> _______________________________________________
>>> srslte-users mailing list
>>> srslte-users at lists.softwareradiosystems.com
>>> http://www.softwareradiosystems.com/mailman/listinfo/srslte-users
>>>
>> _______________________________________________
>> srslte-users mailing list
>> srslte-users at lists.softwareradiosystems.com
>> http://www.softwareradiosystems.com/mailman/listinfo/srslte-users
>>
>
> _______________________________________________
> srslte-users mailing list
> srslte-users at lists.softwareradiosystems.com
> http://www.softwareradiosystems.com/mailman/listinfo/srslte-users